On Wed, Jul 2, 2025, at 07:33, Yonghong Song wrote: > Arnd Bergmann reported an issue ([1]) where clang compiler (less than > llvm18) may trigger an error where the stack frame size exceeds the > limit. > I can reproduce the error like below: > kernel/bpf/verifier.c:24491:5: error: stack frame size (2552) exceeds > limit (1280) in 'bpf_check' > [-Werror,-Wframe-larger-than] > kernel/bpf/verifier.c:19921:12: error: stack frame size (1368) > exceeds limit (1280) in 'do_check' > [-Werror,-Wframe-larger-than] > > Use env->insn_buf for bpf insns instead of putting these insns on the > stack. This can resolve the above 'bpf_check' error. The 'do_check' error > will be resolved in the next patch. > > [1] https://lore.kernel.org/bpf/20250620113846.3950478-1-arnd@xxxxxxxxxx/ > > Reported-by: Arnd Bergmann <arnd@xxxxxxxxxx> > Signed-off-by: Yonghong Song <yonghong.song@xxxxxxxxx> I have confirmed that the fix addresses the issue in bpf_check(). In the worst case I see on an arm64 (clang-15, kasan) randconfigs, the bpf_stack usage goes down from 1680 bytes to 1024. On powerpc64 (clang-15, allmodconfig, kasan), I see an even larger reduction from from 2112 bytes to 1200 for bpf_check(). Tested-by: Arnd Bergmann <arnd@xxxxxxxx> However, I still see 1952 bytes used in do_check() on powerpc64 allmodconfig. This number is much lower in newer clang versions, I see 1888 bytes for clang-19 but only 1232 bytes for clang-20 and -21. Roughly a third of the 1952 bytes seems to come from each one of do_check_insn() and is_state_visited(), but I haven't found where exactly the stack is consumed there. This may be a powerpc specific issue since on arm64 the same function needs less than 1KB stack space. I also tried turning off individual sanitizers (kasan, array-bounts, trace-pc, ...) and they each seem to have a notable impact on the total stack usage of powerpc64 do_check(), i.e. it's not just a problem triggered by kasan or one particular other option. Arnd
